An analysis of analysts' anchoring behavior

61 Pages·2011·0.94 MB·English

by  Benjamin Oomen

An analysis of analysts’ anchoring behavior

Master thesis financial economics
Erasmus University Rotterdam
Supervisor: Dr. A. P. Markiewicz
By: Benjamin Oomen
Student number: 299353
Abstract

This thesis carries out a research into the behavior of security analysts. The main aim is to
find if these analysts use anchors for their earnings per share predictions, and if so, which
anchor they use. Two proposed anchors are tested; the prior year earnings per share and the
consensus of the first three forecasts made after the announcement of the prior year earnings
per share. I find evidence that the prior year earnings per share as anchor, especially when the
change between the actual earnings per share and the prior year earnings per share is positive.
The consensus of the first three forecasts is used more when the change in earnings per share
is negative. This thesis does not find a conclusive answer to which anchor is used most.

2
Master thesis by Benjamin Oomen, An analysis of analysts’ anchoring behavior
Table
 of
 contents

Abstract
 ..........................................................................................................................................
 2

1
 Introduction
 ..............................................................................................................................
 4

2
 Literature
 review
 ....................................................................................................................
 9

2.1
 Anchoring
 ............................................................................................................................................
 9

2.2
 Analysts’
 forecast
 literature
 ......................................................................................................
 11

2.2.1
 Strategic
 bias
 ...........................................................................................................................................
 12

2.2.2
 Optimistic
 bias
 ........................................................................................................................................
 13

2.2.2
 Herding
 bias
 ............................................................................................................................................
 14

2.2.3
 Analysts’
 accuracy
 and
 characteristics
 ........................................................................................
 17

2.3
 Anchoring
 in
 analysts’
 forecasts
 .............................................................................................
 22

2.4
 Conclusions
 on
 literature
 ...........................................................................................................
 23

3
 Methodology
 ..........................................................................................................................
 24

3.1
 Methodology
 idea
 ..........................................................................................................................
 24

3.2
 The
 dependent
 variables
 ............................................................................................................
 25

3.3
 The
 independent
 variables
 ........................................................................................................
 26

3.4
 The
 regression
 ................................................................................................................................
 27

Where:
 ...........................................................................................................................................
 29

4
 Data
 ...........................................................................................................................................
 31

5
 Emperical
 results
 .................................................................................................................
 36

5.1
 Results
 prior
 year
 EPS
 anchor
 .................................................................................................
 36

5.2
 Regression
 results
 consensus
 EPS
 anchor
 ..........................................................................
 44

5.3
 Comparison
 between
 forecasts
 anchored
 on
 the
 prior
 year
 EPS,
 and
 forecasts

anchored
 on
 the
 first
 consensus
 of
 three
 forecasts
 ................................................................
 50

6
 Conclusions
 ............................................................................................................................
 52

7
 References
 ..............................................................................................................................
 54

8
 Appendix
 .................................................................................................................................
 57

3
Master thesis by Benjamin Oomen, An analysis of analysts’ anchoring behavior
1
 Introduction

The field of behavioral finance has always interested me, because I think the only true way to
perceive the behavior of mankind can be achieved through seeing mankind as human.
Although this might sound odd, most courses I have completed during my studies have
assumed mankind in financial markets as completely rational beings. It is only human to
make mistakes, and what might seem irrational might be true but on the other hand what
might seem true can be irrational. Still a lot of the theories that assume human rationality are
used today. When the opportunity came along to conduct research that questioned the
complete rationality of the best forecasts in the market, I gladly accepted. The only way the
field of finance can move forward is to stop assuming rationality, and this is an opportunity to
help. The subject of anchoring is attractive because I think it is a bias almost everybody uses
to make an estimation.

Behavioral finance
One of the first articles where psychology meets economics was written by: Daniel
Kahneman and Amos Tversky in 1974. In their paper about judgment under uncertainty, they
address multiple heuristics in human behavior that lead to biases that should not be able to
exist under theories that assume rationality. One of the heuristics they present is anchoring
and adjustment. Their definition comes down to this: a person is likely to use an anchor, when
he needs to make an estimate of the occurrence of a phenomenon. He then adjusts his
estimate from this anchor, but does not adjust sufficiently. This will result in an estimate that
will fall between the anchor and the real value. In other words, people tend to overuse their
anchor, which results in an underuse of other information.
Analyst literature
Quite some literature has been written about analyst forecasts and why these are inaccurate.
Explanations were found for some systematic errors, which can be caused by intentional
biases as well as non-intentional biases. It seems that as these analysts have a job that forces
them to perceive the markets and make forecasts, they are excellent subjects to test for biases
in their behavior. They can be called experts, as they devote a lot of time to produce these
forecasts. Also there is evidence that the magnitude of earnings per share forecast errors could
lead to job termination (Hong, Kubik, & Solomon, 2000), so they have a clear incentive to be
accurate.
4
Master thesis by Benjamin Oomen, An analysis of analysts’ anchoring behavior
Hypotheses
The earnings per share (EPS from this point forth) forecast of an analyst is a good proxy for
an expert opinion. If I are able to find that a heuristic is used in the forecasts of these analysts,
I can conclude that even these agents do not live up to the assumption of rationality. The main
focus of this thesis will be to see if the EPS forecasts of analyst are subject to the heuristic
called anchoring and adjustment and what situations. It becomes easier to forecast the EPS for
a given year when the first three quarters for that given year are known. To establish an
environment where analysts are prone to using anchors, the forecasts have to be made in a
time that there is not much certainty about the EPS. Thus only the forecasts made in the first
three months are going to be tested. This is defined in the following hypothesis:
H1:
 Analysts
 use
 anchors
 for
 their
 EPS
 forecasts
 in
 the
 first
 three
 months
 after
 the
 earnings

announcement
 under
 specific
 circumstances
 and
 fail
 to
 adjust
 sufficiently

The question that arises is what anchors these analysts would use. The anchor used should be
a valid anchor and a relevant one at the time when the forecast is made. In the first three
months after the announcement of the EPS, the prior year EPS is an anchor that is likely to be
used. Therefore I expect that the analysts will use the prior year EPS as an anchor. In this
thesis the emphasis will be on the properties of anchoring. The main goal of this thesis is to
establish which analysts will express the heuristic of anchoring and adjustment, why and for
which companies. Hong Kubik and Solomon (2000), as well as Mikhail, Walther and Willis
(1997) show that experienced and inexperienced forecasters forecast differently. Because I
expect experienced analysts to be better at forecasting I expect them to display the heuristic of
anchoring less often. Therefore the following hypotheses will be tested:

 H2a:
 Analysts
 use
 the
 prior
 year
 EPS
 as
 an
 anchor
 if
 they
 are
 inexperienced

As the volatility of company earnings becomes bigger, there will be more uncertainty about
what the next EPS will be. Therefore I expect that there will be a positive relationship
between the volatility of earnings for a company and the use of anchors. This is defined in the
following hypothesis:
H2b:
 Analysts
 use
 the
 prior
 year
 EPS
 as
 an
 anchor
 if
 the
 volatility
 of
 the
 company
 is
 bigger

5
Master thesis by Benjamin Oomen, An analysis of analysts’ anchoring behavior
When companies are smaller, less is known about the company, because there normally is
less analyst attention. For smaller companies it will be harder to forecast the EPS. My
expectation is therefore that the size of the company has a negative influence on the behavior
of anchoring. This is defined in the following hypothesis:
H2c:
 Company
 size
 has
 a
 negative
 influence
 on
 the
 behavior
 of
 anchoring

Das et al. (1998) and Easterwood and Nutt (1999) show that systematically analysts’ forecasts
are too optimistic. Analysts might only anchor then when the change in EPS from the prior
year to the actual EPS is negative. Considering the optimism found in earlier papers, the
expectation is that analysts use the prior year EPS more often as an anchor when the company
produces a smaller EPS than the prior year. This is defined in the following hypothesis.
H2d:
 Analysts
 use
 the
 prior
 year
 EPS
 as
 an
 anchor
 if
 the
 company
 produces
 a
 smaller
 EPS

than
 the
 year
 before

In literature concerning analysts, much attention has been dedicated to the evidence found
that suggests that analysts herd. Herding means that analysts make forecasts close to a
consensus. The consensus forecast could be seen as an average of the forecasts that are made
by other analysts. In research, by Hong Kubik and Solomon (2000), proof has been found that
herding is a consequence of career concerns. The authors find a theoretical basis for herding,
because especially inexperienced analysts are more likely to be terminated when they make
bold forecasts. As herding is a well-established phenomenon, the second anchor that analysts
use might therefore be the first consensus of three forecasts. Evidence has been found that
inexperienced analysts herd more often, so I expect to find that the inexperienced analysts
also anchor more on the first three forecasts. This leads to the following hypothesis:
H3a:
 Analysts
 anchor
 on
 the
 consensus
 of
 the
 first
 three
 forecasts
 if
 they
 are
 inexperienced

The expectation is that when the volatility of earnings becomes bigger for a company, the job
of forecasting the EPS will become harder. Therefore a test will be executed if the analysts
anchor more when the earnings volatility becomes bigger. To test this property of anchoring,
the following hypothesis is defined:
H3b:
 Analysts
 anchor
 on
 the
 consensus
 of
 the
 first
 three
 forecasts
 if
 the
 volatility
 of
 the

earnings
 of
 a
 company
 is
 bigger

6
Master thesis by Benjamin Oomen, An analysis of analysts’ anchoring behavior
Company size is expected to have a negative influence on the heuristics of anchoring, because
smaller companies will be harder to forecast the EPS for. Less is known about smaller
companies because there is less analyst attention for these companies. The following
hypothesis is therefore defined:
H3c:
 There
 is
 a
 negative
 relation
 between
 the
 company
 size
 and
 the
 amount
 of
 anchored

forecasts
 on
 the
 consensus
 of
 the
 first
 three
 forecasts

As discussed above, analysts have the tendency to be too optimistic in their forecasts (Das et
al (1998), Easterwood and Nutt (1999)). Therefore I expect that analysts will anchor more on
the first three forecasts when the direction of the earnings change is negative. This is defined
in the following hypothesis:
H3d:
 Analysts
 anchor
 on
 the
 consensus
 of
 the
 first
 three
 forecasts
 if
 the
 company
 produces

a
 smaller
 EPS
 than
 the
 prior
 year.

The two different anchors are going to be tested with two regressions. The forecasts that are
made using an anchor are going to be coded with a one; the forecasts that have not used an
anchor are going to be coded with a zero. The anchor variables are going to be the dependent
variable. With a binary regression the effect of the dependent variables experience, company
size, earnings volatility and negative earnings change will be determined. To control for the
year 1999, there will also be a dummy for this year. The testing period runs from 1993 to
2003. The data over these years will be pooled, so an overall measure of the effects of the
independent variables will be measured.
Relevance
At this point not much has been written about anchoring in the analyst literature. Furthermore,
only testing for the first three months after the announcement of prior year EPS is an
approach, which has not yet been used in the literature concerning analyst behavior. The
anchor proposed in this thesis, using the first three forecasts after the announcement of the
EPS combines the herding literature with the anchoring literature. This subject has barely
been tackled in economic literature.
In the next chapter the literature discussed in short above will be treated more extensively.
Chapter three will contain the methodology. The fourth chapter will cover the data
7
Master thesis by Benjamin Oomen, An analysis of analysts’ anchoring behavior
transformation, and descriptive statistics. In the fifth chapter the regression results, and tables
of the econometric analysis will be presented. This will be accompanied by a discussion and
an economic interpretation of the results. In the sixth and last chapter this thesis will be
summarized.

8
Master thesis by Benjamin Oomen, An analysis of analysts’ anchoring behavior
2
 Literature
 review

The field of behavioral finance consists of two different sciences, psychology and finance.
Therefore the literature of this thesis will discuss papers from both sciences. First the papers
about anchoring will be covered, to discuss which properties have been found in psychology
literature and economic literature. Next literature about analysts will be treated. The analyst
literature is split up in different sections because there are a lot of different perceptions on
why analysts make errors. First there will be an intro into the strategic bias of forecasts.
Subsequently I will discuss two different strategic biases in depth. First there will be a
paragraph on optimistic bias, next there will be a paragraph on herding bias. The paragraph
after this will discuss different papers that have tested analyst characteristics and their
influence on accuracy of analysts. The following subchapter will discuss the papers that
covered analysts and anchoring. The last subchapter will give a small summary of the whole
chapter.

2.1
 Anchoring

Over 37 years ago, Kahneman en Tversky (1974) published a paper that described several
behavioral heuristics and biases. The three heuristics they discuss are ‘representativeness’,
‘availability’ and ‘anchoring and adjustment’. They define anchoring and adjustment as a way
of underestimating new information and overestimating the chosen anchor. If someone has to
make a forecast, say the number of people living in the second city of France, they choose a
point with some relevance (for instance the amount of people living in Paris) and then adjust
from that point and make an estimate. The shortcoming that people show is that the estimate
is typically not adjusted enough, thus leading to a number that will typically be between the
anchor and the real value. For the example concerning the number of people living in the
second city of France, Kahneman and Tversky predict that, when given the number of people
living in Paris, people will overstate the number of people living in the second city of France.
The properties of this heuristic have been tested more than once in different psychological
and economic papers. Northcraft and Neale (1987) found some very convincing evidence by
conducting a laboratory experiment (or field experiment). They tested if two different
subgroups, real estate agents and a group of students, display the bias of anchoring and
adjustment. The setup is that they see if a listing price of a house will influence their opinion
9
Master thesis by Benjamin Oomen, An analysis of analysts’ anchoring behavior
on the price of this house. The individuals get a document containing ten pages of information
on the house. The information is the same for everyone, except for the listing price. Apart
from the information that they receive, they are also driven around the house to see the
neighborhood and the surroundings of the house. The results show that the anchor has a
significant influence on the price tag they put on the house. The results are the same for
different subgroups, and even if the anchor is not realistic at all, it still has a big influence.
We can thus conclude that the experts, who did not admit using the anchor, as well as the
students with not so much experience in valuing houses both display this behavioral
distinction.
Adding to the evidence that experts might display anchoring and adjustment, Wright and
Anderson (1989) constructed another field experiment. They view anchoring and adjustment
as a two-staged phenomenon. First people tend to remember a previous number (the anchor)
and then adjust but fail to do so appropriately. They expect that if they increase the situation-
familiarity, that the part of unintended anchoring disappears. In that way subjects might skip
stage one of the estimation process and make an estimation based on new information. Again,
the subjects display the bias. The subjects are asked if they think the chance on a phenomenon
is bigger (smaller) than 0.25 (0.75) and how much bigger (smaller). In both the familiar
situations to the subject (i.e. an phenomenon in a field they know) and the unfamiliar
situations, the chances given, i.e. 0.25 and 0.75, have a big influence on their estimate. Again
this can also be seen as evidence that experts also use anchors.
Adding to the fact of anchor relevance, Whyte and Sebenius (1997) test if multiple anchors
have a decreasing effect on the anchoring bias. According to the authors, there is reason to
believe that multiple anchors, somewhat creating confusion, might decrease the effect of
anchoring and lead to better estimates. Their results show that the first number the subjects
see has a significant influence. Even after they have been presented with more relevant
numbers later on, the first number they see still has an effect. Interestingly they documented
the same conclusion as Northcraft and Neale (1987) that even when the number does not have
any relevance at all, the anchor still has an influence. They even go as far as including a
number in a document and tell the subjects that this number is a typing error. The conclusion
that can be induced from this research is that any number could be of influence on the
anchoring of human beings. This research suggests that there are numerous possible anchors
for analysts from which they can depart to make their forecast.
One would assume that groups would not be subject to anchoring, but this does not hold
either. Ritov (1996) tested if, in negotiations, anchoring will also influence the price the
10
Master thesis by Benjamin Oomen, An analysis of analysts’ anchoring behavior